BEGIN:VCALENDAR VERSION:2.0 PRODID:-//CERN//INDICO//EN BEGIN:VEVENT SUMMARY:Empowering South African Research through the Altron AI Factory DTSTART;VALUE=DATE-TIME:20251203T094000Z DTEND;VALUE=DATE-TIME:20251203T100000Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-3054@events.chpc.ac.za DESCRIPTION:Speakers: Bongani Andy Mabaso (ALTRON)\nThe Altron AI Factory provides South Africa’s academic community with secure\, locally hosted access to enterprise-grade AI infrastructure and services. Built in partne rship with NVIDIA and hosted in Teraco’s AI-ready data centre\, it offer s GPU-as-a-Service and AI-as-a-Service to accelerate research without heav y infrastructure costs.\nThis presentation highlights how universities and research institutions can leverage the AI Factory to advance data-driven studies\, maintain data sovereignty\, and collaborate across disciplines. Through high-performance computing\, curated models\, and managed services \, the Altron AI Factory bridges the gap between academic research and ind ustrial innovation to partner with South African academia to lead in the A I era.\n\nhttps://events.chpc.ac.za/event/155/contributions/3054/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/3054/ END:VEVENT BEGIN:VEVENT SUMMARY:Hybrid Stacking and Embedded Regression with Multi-Phase Feature S election for Explainable Crop Yield Prediction in Botswana DTSTART;VALUE=DATE-TIME:20251203T133500Z DTEND;VALUE=DATE-TIME:20251203T135000Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-2959@events.chpc.ac.za DESCRIPTION:Speakers: Kalu Ubi Kalu (University of Botswana)\, George And erson (Co-Author)\, Audrey Masizana (University of Botswana)\nHybrid Stac king and Embedded Regression with Multi-Phase Feature Selection for Explai nable Crop Yield Prediction in Botswana \nAbstract \nIn Sub-Saharan Africa 's climate instability\, inaccurate data\, and lack of precision agricultu ral tools make it extremely difficult to predict crop yields with any degr ee of accuracy. These restrictions are especially critical in Botswana\, w here most agricultural activities are rain-fed and highly vulnerable to en vironmental changes. To provide accurate\, comprehensible\, and context-sp ecific yield predictions for four staple crops: Maize\, Millet\, Pulses\, and Sorghum. This study uses a hybrid machine learning approach. The appro ach integrates multiple regression algorithms: Random Forest\, XGBoost\, S upport Vector Regression\, and Multi-Layer Perceptron within a stacked ens emble architecture tailored to Botswana’s agricultural data context. To optimize predictive power and interpretability\, a multi-phase feature sel ection strategy was applied\, combining entropy filtering\, mutual informa tion\, recursive feature elimination (RFE)\, and engineered temporal featu res through lag variables. \nThis process refined input variables for both the staging models and region-specific selection\, ensuring robust model generalization. Model performance was evaluated using historical yield\, m eteorological\, and soil datasets\, with R²\, RMSE\, and MAE employed as metrics. The Stacking Hybrid Regression Model performed exceptionally well in yield prediction for pulses and sorghum\, achieving the best performan ce with R2 = 0.94\, RMSE = 0.60 t/ha\, and MAE = 0.32 t/ha. The most signi ficant predictors were rainfall\, temperature fluctuation\, and lagged yie ld values\, according to a unified interpretability framework that was pro duced by combining SHapley Additive exPlanations (SHAP) with entropy analy sis. Surprisingly\, entropy research showed that Sorghum had a greater pre dictor complexity and shown the ability to adjust to unpredictable weather . Time-horizon stability of the model was confirmed by forward simulations for 2025–2028. \nThese results confirm that interpretable hybrid ensemb les can satisfy precision agriculture's accuracy and transparency requirem ents when reinforced by multi-phase feature selection. The suggested appro ach supports climate risk management tactics for Botswana's farmers by pro viding useful information for early-season production projection and input distribution. Additionally\, other sub-Saharan regions with comparable en vironmental and data-related constraints may find the methodology applicab le. \nKeywords: predictive crop yield\, precision agriculture\, Botswana\, XAI\, multi-phase feature selection\, hybrid ensemble models\, and SHAP.\ n\nhttps://events.chpc.ac.za/event/155/contributions/2959/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/2959/ END:VEVENT BEGIN:VEVENT SUMMARY:Recognising South African Voices: A Multilingual ASR Pipeline DTSTART;VALUE=DATE-TIME:20251203T142000Z DTEND;VALUE=DATE-TIME:20251203T143500Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-3048@events.chpc.ac.za DESCRIPTION:Speakers: Mahlatse Mbooi (CSIR)\nSouth Africa’s rich linguis tic diversity poses unique challenges for artificial intelligence systems\ , particularly in automatic speech recognition (ASR) where multilingual sp eakers frequently switch languages mid-conversation. This study proposes a robust ASR pipeline tailored for code-switched speech in health settings\ , addressing practical issues such as overlapping dialogue\, background no ise\, and inconsistent language usage. The pipeline will integrate multili ngual acoustic models and language-specific preprocessing techniques\, tra ined on a standardised dataset comprising South African languages includin g isiZulu\, Sepedi and English.\nBy focusing on pipeline design\, dataset standardisation and multilingual integration\, this work demonstrates how AI can be built to truly understand South African voices rather than ignor ing them. Structured and reproducible approaches to code-switched data lay the foundation for inclusive\, fair\, and context-aware AI that represent s local language communities and highlight the broader opportunities for l everaging multilingual data responsibly.\n\nhttps://events.chpc.ac.za/even t/155/contributions/3048/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/3048/ END:VEVENT BEGIN:VEVENT SUMMARY:Q & A DTSTART;VALUE=DATE-TIME:20251203T145000Z DTEND;VALUE=DATE-TIME:20251203T150500Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-3047@events.chpc.ac.za DESCRIPTION:Speakers: Q & A ()\nQ & A\n\nhttps://events.chpc.ac.za/event/1 55/contributions/3047/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/3047/ END:VEVENT BEGIN:VEVENT SUMMARY:Factors associated with sexually transmitted infection literacy am ong men who have sex with men and transgender people in Soweto: A machine learning approach DTSTART;VALUE=DATE-TIME:20251203T143500Z DTEND;VALUE=DATE-TIME:20251203T145000Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-2964@events.chpc.ac.za DESCRIPTION:Speakers: Claris Siyamayambo (University of Johannesburg)\nSex ually transmitted infections (STIs) remain a significant public health cha llenge in Sub-Saharan Africa (SSA)\, particularly among key populations su ch as men who have sex with men (MSM) and transgender individuals. This st udy aimed to assess the level of STI literacy within this population\, ide ntify its demographic\, behavioral\, and structural predictors\, and explo re its influence on knowledge\, attitudes\, behaviors\, and healthcare-see king. A retrospective observational mixed-methods approach was employed\, combining logistic regression\, structural equation modeling (SEM)\, and e xplainable machine learning (SHAP) to analyze data collected from 1\,240 M SM and transgender individuals in Soweto\, South Africa. The main outcome variable\, STI literacy\, was operationalized both as a composite score (b inary: high and low) and as a categorical label (1 and 0)\, enabling both inferential and predictive modeling. Results revealed that 28.1% of parti cipants demonstrated adequate STI literacy. Key positive predictors includ ed younger age\, prior STI testing\, higher education\, being single or ma rried\, female gender identity\, and personal STI history. In contrast\, o lder age\, unemployment\, lower education\, substance use\, and frequent s exual activity were associated with lower literacy. Structural equation mo deling illuminated how STI testing experience acts as a cue to action\, wh ile stigma\, cost\, and fear serve as barriers. SHAP analysis confirmed th ese insights\, highlighting modifiable predictors such as information-seek ing\, communication confidence\, and testing accessibility. The study's fi ndings were interpreted through Nutbeam’s Health Literacy Framework\, th e Health Belief Model (HBM)\, and the Theory of Planned Behavior (TPB). Th ese frameworks helped contextualize the behavioral pathways linking sociod emographic factors to STI literacy and preventive actions. Notably\, TPB c onstructs such as subjective norms and perceived behavioral control were p articularly influential. This study contributes to the STI prevention lite rature by quantifying literacy gaps\, modeling predictive pathways\, and d emonstrating how behavioral theory and machine learning can inform targete d interventions. It recommends multi-level approaches that go beyond aware ness to address stigma\, build self-efficacy\, and enhance access to sexua l health services. These insights are vital for designing inclusive\, theo ry-driven public health strategies in SSA.\n\nhttps://events.chpc.ac.za/ev ent/155/contributions/2964/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/2964/ END:VEVENT BEGIN:VEVENT SUMMARY:SADiLaR's repository as a cyber-infrastructure: Improving data acc essibility for South African languages DTSTART;VALUE=DATE-TIME:20251203T140500Z DTEND;VALUE=DATE-TIME:20251203T142000Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-2963@events.chpc.ac.za DESCRIPTION:Speakers: Menno van Zaanen (South African Centre for Digital L anguage Resources)\nComputing processes typically require input data to pe rform actions\nthat generate output data. While input data can sometimes b e generated\ncomputationally\, it often originates from external sources. In Natural\nLanguage Processing and Digital Humanities\, this input is typ ically\nsourced from human activities\, including spoken or written langua ge\nand music.\n\nIn the current era of Large Language Models (LLMs) that provide\npractically usable tools\, access to appropriate training data is \nessential. These models generally perform better when larger data\ncolle ctions are available for training\, making data accessibility\ncrucial.\n\ nFor most South African languages\, only limited amounts of digitally\nacc essible data are available. Many existing data collections are\nsourced fr om government websites\, providing texts in highly specific\ngenres. Texts from diverse genres --- including newspaper articles\,\nliterary works\, and social media data --- are not openly accessible in\ndigital formats.\n \nThe SADiLaR (South African Centre for Digital Language Resources)\nrepos itory hosts data collections that are as openly accessible as\npossible. T he repository currently contains 357 directly downloadable\nitems and 56 m etadata-only items (indicating the existence of data\ncollections).\n\nThe underlying principle of SADiLaR's repository is that providing a\ncentral ized space for data collections makes them more easily findable\nand acces sible. (Additionally\, submitted data collections are\nrequested to be as interoperable and reusable as possible to ensure\nadherence to FAIR princ iples.)\n\nThis abstract serves as a call for action with two main objecti ves.\nFirst\, we encourage researchers to submit their digital language da ta\nto the SADiLaR repository. Contributing to the repository increases\nt he availability of South African language data\, making it more easily\nfi ndable and accessible. This data can then be used for training models\,\nu ltimately benefiting language users\, for example\, through the\ndevelopme nt of LLMs for these languages. (Note: copyright remains with\nthe origina l copyright owner\; contributing to the repository does not\ntransfer copy right ownership.)\n\nEach contribution to SADiLaR's repository receives a persistent\nidentifier\, enabling consistent referencing of data collectio ns. These\nidentifiers can be used as citations in publications\, ultimate ly\nbenefiting researchers associated with the data collections.\n\nSecond \, we encourage researchers to utilize the data collections\navailable in SADiLaR's repository. The repository contains a wealth of\nuseful data col lections\, and searching there first can streamline\nresearch processes. S ADiLaR's repository exists to facilitate work in\nNatural Language Process ing and Digital Humanities\; collectively\, we\ncan leverage this cyber-in frastructure to advance our fields.\n\nhttps://events.chpc.ac.za/event/155 /contributions/2963/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/2963/ END:VEVENT BEGIN:VEVENT SUMMARY:Computational Model for Identifying Sepitori in Multilingual Texts DTSTART;VALUE=DATE-TIME:20251203T135000Z DTEND;VALUE=DATE-TIME:20251203T140500Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-2961@events.chpc.ac.za DESCRIPTION:Speakers: Dan Masethe (Tshwane University of Technology)\nThe Sepitori language (also known as Pitori or Pretoria Sotho) is a dynamic an d evolving creole language predominantly spoken in urban townships of Pret oria\, South Africa. It blends Setswana\, Sesotho\, Afrikaans\, and Englis h\, with frequent instances of code-switching and slang. Despite its wides pread usage\, Sepitori remains underrepresented in natural language proces sing (NLP) tasks\, particularly in language identification and text proces sing.\n\nThis paper proposes the development of a Sepitori Language Identi fication (ID) Model\, designed to classify and distinguish Sepitori text f rom other South African languages. The model addresses the unique challeng es of multi-language mixing\, informal vocabulary\, and varying dialects w ithin the Sepitori speech community. By leveraging machine learning techni ques and deep learning models\, including convolutional neural networks (C NN) and transformer-based models (e.g.\, BERT)\, the model utilizes a larg e-scale corpus of annotated Sepitori\, Setswana\, Sesotho\, Afrikaans\, an d English samples. The model incorporates multiple linguistic features\, s uch as n-grams\, word embeddings\, and syntactic patterns\, to accurately identify Sepitori text\, even when it involves heavy code-switching or sla ng.\n\nThis work contributes to the linguistic field by providing a novel computational tool for processing Sepitori\, enabling the automatic detect ion of Sepitori in a variety of contexts\, including social media\, web sc raping\, and corpus development. It also lays the foundation for improving language resources for underserved African languages\, with potential app lications in speech recognition\, machine translation\, and sentiment anal ysis. The model is expected to improve the accessibility and representatio n of Sepitori in digital and computational platforms\, fostering greater i nclusivity for African language speakers in the digital age.\n\nhttps://ev ents.chpc.ac.za/event/155/contributions/2961/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/2961/ END:VEVENT BEGIN:VEVENT SUMMARY:Advancing South Africa’s Research Data Ecosystem: Infrastructure \, Innovation\, and Collaboration. DTSTART;VALUE=DATE-TIME:20251203T090000Z DTEND;VALUE=DATE-TIME:20251203T092000Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-2970@events.chpc.ac.za DESCRIPTION:Speakers: More Manda (CSIR)\nDIRISA: Update on Data Infrastruc ture for Research Data Management and Collaborations\; \nDr More Manda.\n\ nhttps://events.chpc.ac.za/event/155/contributions/2970/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/2970/ END:VEVENT BEGIN:VEVENT SUMMARY:Catalysing Data Centre Investment for Africa’s High-Performance Computing and Climate Action through the Digital Investment Facility (DIF) DTSTART;VALUE=DATE-TIME:20251203T113000Z DTEND;VALUE=DATE-TIME:20251203T130000Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-2962@events.chpc.ac.za DESCRIPTION:Speakers: Mulalo Mphidi (GIZ)\nThe rapid growth of Africa’s data-intensive research\, artificial intelligence (AI)\, and high-perform ance computing (HPC) workloads is driving unprecedented demand for resilie nt and sustainable data infrastructure. Data centres are emerging as criti cal enablers of scientific discovery\, cloud adoption\, and digital innova tion\, yet the region continues to face significant barriers: limited loca l hosting capacity\, reliance on international facilities\, high latency\, data sovereignty concerns\, and a shortage of investment-ready projects.\ nThe Digital Investment Facility (DIF)—a Team Europe initiative co-funde d by the European Commission\, Germany\, and Finland\, and implemented joi ntly by GIZ and HAUS—addresses these gaps by boosting investment in gree n and secure digital infrastructure\, with a focus on data centres and Int ernet Exchange Points (IXPs). Operating as a project preparation and advis ory facility\, DIF supports projects from early design to contract closing \, enhancing bankability through technical and financial advisory services \, pre-feasibility studies\, ESG integration\, and investor matchmaking.\n Crucially\, DIF embeds a climate nexus at the core of its work. By promoti ng energy-efficient\, renewable-powered data centres and aligning with ISO 50001 energy management standards\, DIF ensures digital infrastructure pr ojects contribute directly to climate action and the implementation of Nat ionally Determined Contributions (NDCs). Greener data centres reduce emiss ions from digital growth\, enhance resilience through disaster recovery ca pacity\, and enable the digital tools required for climate adaptation (e.g .\, climate modelling\, earth observation\, and early warning systems).\nA t CHPC\, DIF will showcase how its approach enables data centres to meet t he demanding requirements of HPC and advanced research—providing low-lat ency access\, high-availability colocation\, and sustainable cloud platfor ms that can host scientific datasets and AI workloads. The presentation wi ll highlight the emerging pipeline of African digital infrastructure proje cts\, the application of international standards\, and the opportunities f or researchers\, policymakers\, and investors to collaborate in building a digitally sovereign and climate-aligned HPC ecosystem in Africa.\n\nhttps ://events.chpc.ac.za/event/155/contributions/2962/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/2962/ END:VEVENT BEGIN:VEVENT SUMMARY:Q&A DTSTART;VALUE=DATE-TIME:20251203T102000Z DTEND;VALUE=DATE-TIME:20251203T103000Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-2590@events.chpc.ac.za DESCRIPTION:https://events.chpc.ac.za/event/155/contributions/2590/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/2590/ END:VEVENT BEGIN:VEVENT SUMMARY:AI-Assisted Optimization of Large-Scale Climate Data Transfers in South African Research Infrastructure DTSTART;VALUE=DATE-TIME:20251203T100000Z DTEND;VALUE=DATE-TIME:20251203T102000Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-2960@events.chpc.ac.za DESCRIPTION:Speakers: Jonathan Padavatan (University of the Wiitwatersrand )\n# AI-Assisted Optimization of Large-Scale Climate Data Transfers in Sou th African Research Infrastructure\n\n**CHPC Conference 2025\, Cape Town** \n\n## Abstract\n\n### Background and Motivation\n\nThe transfer of large- scale scientific datasets between South African research facilities repres ents a critical bottleneck in computational research workflows. Climate mo deling datasets of the Global Change Instituite\, Wits University \, as od f Aug2025\, are just over 540TB over 3 users\, particularly from the Conf ormal-Cubic Atmospheric Model (CCAM).Optimized transfer strategies between the Centre for High Performance Computing (CHPC) and the Data Intensive R esearch Initiative of South Africa (DIRISA) storage systems are thus neces sary for resilient data flows between HPC\, storage and local analsysi s compute facilities.Current data transfer tools such as Globus Connect id entify bottlenecks within data flow circuits\, however manual command line iRODS interfaces present significant challenges for reliable data transf er through AI-assisted optimization."\n\n### Methodology: AI-Assisted Deve lopment\n\nThis work presents a systematic application of artificial intel ligence tools (Claude Code) to develop filesystem-aware transfer optimizat ion solutions. The AI-assisted development process generated three complem entary tools in under 4 hours of development time:\n\n1. **Performance ben chmarking script** for systematic testing of 24-core Data Transfer Node co nfigurations\n2. **Resilient transfer wrapper** with exponential backoff r etry logic and comprehensive verification\n3. **Lustre-aware optimization engine** that dynamically analyzes filesystem striping patterns and adjust s transfer parameters\n\n### Technical Innovation: Lustre Filesystem Integ ration\n\nThe core innovation lies in automated Lustre striping analysis u sing `lfs getstripe` commands\, coupled with dynamic parameter optimizatio n. The system automatically detects:\n- Stripe counts and sizes for optima l thread allocation\n- Object Storage Target (OST) distributions for concu rrency planning \n- File size patterns for buffer optimization\n- Directo ry structures for efficient batch processing\n\n### Performance Results\n\ n**Test Dataset**: 189TB CCAM climate modeling installation (`ccam_install _20240215`)\n- **Source**: CHPC Lustre filesystem (`/home/jpadavatan/lustr e/`)\n- **Destination**: DIRISA iRODS storage (`/dirisa.ac.za/home/jonatha n.padavatan@wits.ac.za/`)\n\n**Validation Testing** (67MB\, 590 files):\n- **Success Rate**: 100.0% (590/590 files transferred successfully)\n- **Tr ansfer Performance**: 0.95 GB/hour sustained throughput\n- **Reliability** : Zero failed transfers with comprehensive verification\n- **Peak Performa nce**: 10.41 MB/s maximum transfer rate\n- **Optimization**: Automatic 8-t hread\, 64MB buffer configuration\n\n**Scalability Analysis**:\n- **Small datasets** (41-67MB): 100% success rate\, 4-11 MB/s\n- **Medium datasets** (17GB): Structure-preserving transfers completed\n- **Large datasets** (2 0-34TB per directory): Systematic optimization applied\n\n### AI Developme nt Impact\n\nThe AI-assisted approach delivered significant advantages:\n- **Development Speed**: Complete toolchain developed in \n\nhttps://events .chpc.ac.za/event/155/contributions/2960/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/2960/ END:VEVENT BEGIN:VEVENT SUMMARY:Leveraging Cyber-infrastructure for Open Science: DIRISA's Contrib ution to Data-Driven Decision-Making in South Africa. DTSTART;VALUE=DATE-TIME:20251203T092000Z DTEND;VALUE=DATE-TIME:20251203T094000Z DTSTAMP;VALUE=DATE-TIME:20260511T023957Z UID:indico-contribution-662-2965@events.chpc.ac.za DESCRIPTION:Speakers: Phoshoko Katlego (NICIS DIRISA)\nThe transition from raw research data to impactful national decisions relies fundamentally on robust\, accessible\, and strategically managed data and data infrastruct ure. This presentation provides a high-level overview of the foundations o f open science and frames the urgency within the unique South African cont ext. It addresses critical systemic challenges\, including data fragmentat ion and the complex dynamics of data ownership and governance that shape t he research landscape.\nThe core focus of the discussion is DIRISA's strat egic mandate as the key national enabler. The presentation illustrates how DIRISA provides national data platforms\, research data management suppor t\, and services that govern the full data lifecycle—from ingestion to l ong-term preservation and sharing.\nThe presentation aims to demonstrate t he national value of data stewardship: how it effectively bridges the gap between theoretical frameworks and practical\, evidence-based decision-mak ing for national benefit. The presentation concludes by exploring emerging trends and future requirements necessary to fully realize data’s potent ial for South Africa's sustainable development.\n\nhttps://events.chpc.ac. za/event/155/contributions/2965/ LOCATION:Century City Conference Centre 1/1-7 - Room 7 URL:https://events.chpc.ac.za/event/155/contributions/2965/ END:VEVENT END:VCALENDAR